CROP CHARACTERISTIC RESEARCH: G. D. Badhwar
GROWTH AND REFLECTANCE ANALYSIS Johnson Space Center
Houston, TX 77058
Justification, Goals, and Objectives
Much of the early research in remote sensing follows along developing
"spectral signatures" of cover types. It was, however, found that a
signature from an unknown cover class could not be matched to a catalog
value of known cover class. This approach was abandoned and "supervised
classification" schemes followed. These were not efficient and required
extensive training. It has been patently clear that data acquired at a
single time could not separate cover types.
A large portion of the proposed research has concentrated on modeling the
temporal behavior of agricultural crops and on removing the need for any
training data in remote sensing surveys — the key to which is the solution
of the so-called "signature extension" problem.
A clear need to develop spectral estimaters of crop ontogenic stages and
yield has existed even though various correlations have been developed.
Considerable effort in developing techniques to estimate these variables
was devoted to this work.
The need to accurately evaluate existing canopy reflectance model (s),
improve these models, use them to understand the "crop signatures," and
estimate leaf area index was the third objective of the proposed work.
The next section gives a synposis of this research effort.
The technical approach consisted of first developing an accurate model that
would describe the temporal development of various spectral transforms with
time, henceforth called a profile, that would depend primarily on crop
characteristics. It would thus permit features to be extracted from real
spectral data that describes a specific crop and would not depend on external
variables such as row direction, sun zenith angle, the atmospheric state,
etc. Having extracted a very small set of features use the canopy reflectance
model (s) to gain (a) better understanding and limitations of existing canopy
models, (b) use the model (s) to accure a deeper physical understanding of
why these features permit crop separation and develop method(s) of deciding,
a priori, what features would permit crop separation, and (c) explorate the
applicability of these models to natural forest communities.
The current research effort has shewn that the Kauth-Thomas (K-T) greenness,
in the spectral space of both the multispectral spectral scanner or the
thematic mapper, can be described by a model of the form,
/ 2ee \ a ?
p(t) = P 0 + (p m - P 0 ) ( — ) (t - t Q ) Exp [-3(t - t Q ) 2 ] (1)
where p(t) the K-T greenness as a function of time, p m , the maximum value of
greenness reached at time of peak greenness
t p -
p 0 the value of soil greenness at times at and before emergence, t 0 , and a
and 3 are two crop and condition specifics constant. These two constant are
related to the inflection points of the profile
ti - t 0 =
(2a + 1) - v/8a + 1
t 2 - t 0 =
"(2a + 1) +/8a + 1
This research effort has established that the peak greenness above the soil
line, G max = Pm “ Pq> the separation
1 a f
cr = 1 2 “ 1 1 =
— + — [1
- (1 - - )l
2(5 2(5 L
and the time of peak greenness, t p ,'are three characteristics or features
that carry 95% of all information (Fisher information criteria) available
in both spectral and temporal data and has led to a drastic reduction in
the number of variables and a simplification of classifier design.
Figure 1 shows the power of these features in separating two summer crops,
corn/soybeans, based on data extracted from the thematic mapper. The axis
out of the plane of paper is the number of pixels, the other two axes beifK)
G max and 0 days. The distributions are more or less gaussian and provide
excellent separabi 1 i ty . Not only has it been shown that these features (G max ,
o, and tp) are applicable to different sensor systems but are truly
"signature extendable" over vast areas in the United States and, for the
first time ever in remote sensing, to areas in Argentina. In particular,
it has been found that, (i) G max (soybeans) > G max (corn), (ii) o (soybeans) <
a (corn), and (iii) t p (soybeans) > t p (corn).
jsrifW-n*?; s tmzvf
Illo" e f his fill' COr ?; and s °? bea,ls and appears to be true for
,n c S TSI^ f Tc r ^o„.
It has also been shown that integral,
which acts in a manner very similar to the leaf area duration i<? ct-renniw
inThl^Ir . to Jf lel J 1n the case of b oth corn and soybeans. More research*
autom^nv 1r AnH 10 K- 1S f^ al1ed f ° r; however » the Potential of a true integrated
an5°S^„rlre1f«r„mf„ P 9 r | a d |p“ ,0n SySt “ ipp,i “ b,e tp S'° bal “rn ’
In order to understand the reasoning behind why (soybeans! > a
aI«ost ooavorsely, the existing canopy reflecting Sodl°f llle ose^Villll’
llflll nX^e'l™ Z]‘ I? £”*
le r |a n |a f bm“ d ca 0 n Pt |riaulla7ld?a n “''i?!.' 0 Pr ° ,tde * <* '
The 'exf sting canopy reflectance models'had not been subiected fn a
.. ’s , d CUPID canopy reflectance models were very carefully evaluated for
e 0 ‘° rn and soybeans. AH of these models captured the s!lllnt
l!fSn 0 r he Canopy reflec tance, with the SAIL model showing the best
overall performance. However, it was found that it is necessary to include
u e U !l , r , ° n ? 1 characteristics of leaves into these Sde?s
iMlIJinn'if ? *5 perf ? rmance of the SAIL model before and afte- the
inclusion °f leaf specular reflectance systematic angular defic encies of
of th^Io^s! 0 ^^^;^^ KTJfSj;nyIi h «i s S^5T v SuS e .r ?!Si n :s.
— Trr :l r °‘ '
£ II r ilaqir^oll-: lllrilrfl^cl-ln'^l^irj??! d “
(1). The meaning of the t] , t p , and a would of course be verj - different.
(1) For the first time in remote sensing, crop features have been found
that are truly "signature extendable." In case of corn and soybeans
they have been shown to be applicable to vast geographic regions of
the United States for four years 1978, 1979, 1980, and 1982 and are
extendable to Argentina. Based on SAIL canopy reflectance model, the
reasons for this applicability have been understood.
(2) It has been shown that critical ontogenetic crop stages can be estimated
from spectral data. Based on this work and more detailed work, it is
suggested that it may be more accurate to estimate crop phenology
using spectral data than current methods.
(3) Preliminary evidence suggests that the area of greenness profile from '
t] (when new leaf development stops) to t2 (dent) is strongly correlated
The above three results have made it possible to seriously consider an
automatic and objective crop production system.
(4) An improved canopy reflectance model has been developed that includes
the leaf specular component.
(5) The effectiveness of these models in estimation of leaf area index
of wheat, corn, and soybeans and recently in study of forest species
separation and aspen leaf area estimation has been demonstrated.
Most of the current work on feature extraction, ontogenetic stage and yield
has been done on corn and soybean. Some start was made on wheat, barley,
and oats. The work on spring grains should be intensified. Currently,
no technique exists to separate the three crops. Additional work on
ontogenetic stage and yield of corn and soybeans still needs to be done.
The performance of existing canopy models on forest canopies has been found
to be sadly lacking, much more so for coniferous forest than for decidous
forest. Major improvements in these models are called for and a corresponding
adequate input data set must be collected. Techniques to estimate leaf
area index or phytomass of vegetation cannot be developed realiably without
such an effort.
cn:c:::/.i r.v; .:
OF POOH: QUALITY